Taking the upstream basin of the Ruoergai Hydrological Station of the Heihe River in the source area of the Yellow River as the study area, a smart Xin’anjiang model coupling the snowmelt soil freeze-thaw module and random forest algorithm was constructed. Runoff simulation was carried out using precipitation, evaporation and temperature data from 2007 to 2019. And the simulation results were compared with the traditional Xin’anjiang model, the Xin’anjiang model coupling snowmelt, soil freeze-thaw and the random forest model to evaluate the adaptability of the smart Xin’anjiang model in the hydrological process simulation in high-cold regions. The results show that the overall simulation ability of the traditional Xin’anjiang model during the verification period is acceptable, but its simulation ability during the melting period is limited and cannot accurately depict the runoff process during the melting period. The simple random forest model can capture the nonlinear characteristics of the hydrological process during the melting period, the overall simulation effect during the verification period is not ideal. The coupling of the snowmelt soil freeze-thaw module increased the uncertainty of the Xin ’anjiang model by 35.6%, but the model stability and simulation performance were good. During the ablation period, the Nash efficiency coefficient